AUTOMATING INTERACTION DYNAMICS NOTATION FOR REAL-TIME TEAM ANALYSIS

Abstract Communication plays a central role in how engineering design teams generate ideas, negotiate decisions, and make progress. Interaction Dynamics Notation (IDN) offers a structured way to study these interactions, but its reliance on manual, post hoc coding limits how widely and quickly it can be used. This paper introduces a system that automatically classifies IDN symbols in real time from live team conversations. The system combines speech transcription with a lightweight language model and a short conversational context window to label interactions as they occur. The approach is evaluated in a professional design sprint and compared against human-coded IDN data. Results show that the real-time system achieves accuracy comparable to human inter-coder reliability and to prior automated methods that operate only after a conversation has ended. Hidden Markov Model analysis further indicates that the AI-coded data captures the same dominant interaction patterns and transitions observed in human coding, while smoothing some finer distinctions. A sensitivity analysis of context window size highlights practical trade-offs between accuracy and latency. Overall, this work makes IDN more accessible for studying team interactions and supports future tools that help teams understand and reflect on their communication as it unfolds.